drivers/vhost/vsock.c

   1 /*
   2  * vhost transport for vsock
   3  *
   4  * Copyright (C) 2013-2015 Red Hat, Inc.
   5  * Author: Asias He <asias@redhat.com>
   6  *         Stefan Hajnoczi <stefanha@redhat.com>
   7  *
   8  * This work is licensed under the terms of the GNU GPL, version 2.
   9  */
  10 #include <linux/miscdevice.h>
  11 #include <linux/atomic.h>
  12 #include <linux/module.h>
  13 #include <linux/mutex.h>
  14 #include <linux/vmalloc.h>
  15 #include <net/sock.h>
  16 #include <linux/virtio_vsock.h>
  17 #include <linux/vhost.h>
  18 #include <linux/hashtable.h>
  19
  20 #include <net/af_vsock.h>
  21 #include "vhost.h"
  22
  23 #define VHOST_VSOCK_DEFAULT_HOST_CID    2
  24 /* Max number of bytes transferred before requeueing the job.
  25  * Using this limit prevents one virtqueue from starving others. */
  26 #define VHOST_VSOCK_WEIGHT 0x80000
  27 /* Max number of packets transferred before requeueing the job.
  28  * Using this limit prevents one virtqueue from starving others with
  29  * small pkts.
  30  */
  31 #define VHOST_VSOCK_PKT_WEIGHT 256
  32
  33 enum {
  34         VHOST_VSOCK_FEATURES = VHOST_FEATURES,
  35 };
  36
  37 /* Used to track all the vhost_vsock instances on the system. */
  38 static DEFINE_SPINLOCK(vhost_vsock_lock);
  39 static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8);
  40
  41 struct vhost_vsock {
  42         struct vhost_dev dev;
  43         struct vhost_virtqueue vqs[2];
  44
  45         /* Link to global vhost_vsock_hash, writes use vhost_vsock_lock */
  46         struct hlist_node hash;
  47
  48         struct vhost_work send_pkt_work;
  49         spinlock_t send_pkt_list_lock;
  50         struct list_head send_pkt_list; /* host->guest pending packets */
  51
  52         atomic_t queued_replies;
  53
  54         u32 guest_cid;
  55 };
  56
  57 static u32 vhost_transport_get_local_cid(void)
  58 {
  59         return VHOST_VSOCK_DEFAULT_HOST_CID;
  60 }
  61
  62 /* Callers that dereference the return value must hold vhost_vsock_lock or the
  63  * RCU read lock.
  64  */
  65 static struct vhost_vsock *vhost_vsock_get(u32 guest_cid)
  66 {
  67         struct vhost_vsock *vsock;
  68
  69         hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) {
  70                 u32 other_cid = vsock->guest_cid;
  71
  72                 /* Skip instances that have no CID yet */
  73                 if (other_cid == 0)
  74                         continue;
  75
  76                 if (other_cid == guest_cid) {
  77                         return vsock;
  78                 }
  79         }
  80
  81         return NULL;
  82 }
  83
  84 static void
  85 vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
  86                             struct vhost_virtqueue *vq)
  87 {
  88         struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
  89         bool added = false;
  90         bool restart_tx = false;
  91
  92         mutex_lock(&vq->mutex);
  93
  94         if (!vq->private_data)
  95                 goto out;
  96
  97         /* Avoid further vmexits, we're already processing the virtqueue */
  98         vhost_disable_notify(&vsock->dev, vq);
  99
 100         for (;;) {
 101                 struct virtio_vsock_pkt *pkt;
 102                 struct iov_iter iov_iter;
 103                 unsigned out, in;
 104                 size_t nbytes;
 105                 size_t len;
 106                 int head;
 107
 108                 spin_lock_bh(&vsock->send_pkt_list_lock);
 109                 if (list_empty(&vsock->send_pkt_list)) {
 110                         spin_unlock_bh(&vsock->send_pkt_list_lock);
 111                         vhost_enable_notify(&vsock->dev, vq);
 112                         break;
 113                 }
 114
 115                 pkt = list_first_entry(&vsock->send_pkt_list,
 116                                        struct virtio_vsock_pkt, list);
 117                 list_del_init(&pkt->list);
 118                 spin_unlock_bh(&vsock->send_pkt_list_lock);
 119
 120                 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
 121                                          &out, &in, NULL, NULL);
 122                 if (head < 0) {
 123                         spin_lock_bh(&vsock->send_pkt_list_lock);
 124                         list_add(&pkt->list, &vsock->send_pkt_list);
 125                         spin_unlock_bh(&vsock->send_pkt_list_lock);
 126                         break;
 127                 }
 128
 129                 if (head == vq->num) {
 130                         spin_lock_bh(&vsock->send_pkt_list_lock);
 131                         list_add(&pkt->list, &vsock->send_pkt_list);
 132                         spin_unlock_bh(&vsock->send_pkt_list_lock);
 133
 134                         /* We cannot finish yet if more buffers snuck in while
 135                          * re-enabling notify.
 136                          */
 137                         if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
 138                                 vhost_disable_notify(&vsock->dev, vq);
 139                                 continue;
 140                         }
 141                         break;
 142                 }
 143
 144                 if (out) {
 145                         virtio_transport_free_pkt(pkt);
 146                         vq_err(vq, "Expected 0 output buffers, got %u\n", out);
 147                         break;
 148                 }
 149
 150                 len = iov_length(&vq->iov[out], in);
 151                 iov_iter_init(&iov_iter, READ, &vq->iov[out], in, len);
 152
 153                 nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
 154                 if (nbytes != sizeof(pkt->hdr)) {
 155                         virtio_transport_free_pkt(pkt);
 156                         vq_err(vq, "Faulted on copying pkt hdr\n");
 157                         break;
 158                 }
 159
 160                 nbytes = copy_to_iter(pkt->buf, pkt->len, &iov_iter);
 161                 if (nbytes != pkt->len) {
 162                         virtio_transport_free_pkt(pkt);
 163                         vq_err(vq, "Faulted on copying pkt buf\n");
 164                         break;
 165                 }
 166
 167                 vhost_add_used(vq, head, sizeof(pkt->hdr) + pkt->len);
 168                 added = true;
 169
 170                 if (pkt->reply) {
 171                         int val;
 172
 173                         val = atomic_dec_return(&vsock->queued_replies);
 174
 175                         /* Do we have resources to resume tx processing? */
 176                         if (val + 1 == tx_vq->num)
 177                                 restart_tx = true;
 178                 }
 179
 180                 virtio_transport_free_pkt(pkt);
 181         }
 182         if (added)
 183                 vhost_signal(&vsock->dev, vq);
 184
 185 out:
 186         mutex_unlock(&vq->mutex);
 187
 188         if (restart_tx)
 189                 vhost_poll_queue(&tx_vq->poll);
 190 }
 191
 192 static void vhost_transport_send_pkt_work(struct vhost_work *work)
 193 {
 194         struct vhost_virtqueue *vq;
 195         struct vhost_vsock *vsock;
 196
 197         vsock = container_of(work, struct vhost_vsock, send_pkt_work);
 198         vq = &vsock->vqs[VSOCK_VQ_RX];
 199
 200         vhost_transport_do_send_pkt(vsock, vq);
 201 }
 202
 203 static int
 204 vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt)
 205 {
 206         struct vhost_vsock *vsock;
 207         struct vhost_virtqueue *vq;
 208         int len = pkt->len;
 209
 210         rcu_read_lock();
 211
 212         /* Find the vhost_vsock according to guest context id  */
 213         vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid));
 214         if (!vsock) {
 215                 rcu_read_unlock();
 216                 virtio_transport_free_pkt(pkt);
 217                 return -ENODEV;
 218         }
 219
 220         vq = &vsock->vqs[VSOCK_VQ_RX];
 221
 222         if (pkt->reply)
 223                 atomic_inc(&vsock->queued_replies);
 224
 225         spin_lock_bh(&vsock->send_pkt_list_lock);
 226         list_add_tail(&pkt->list, &vsock->send_pkt_list);
 227         spin_unlock_bh(&vsock->send_pkt_list_lock);
 228
 229         vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
 230
 231         rcu_read_unlock();
 232         return len;
 233 }
 234
 235 static int
 236 vhost_transport_cancel_pkt(struct vsock_sock *vsk)
 237 {
 238         struct vhost_vsock *vsock;
 239         struct virtio_vsock_pkt *pkt, *n;
 240         int cnt = 0;
 241         int ret = -ENODEV;
 242         LIST_HEAD(freeme);
 243
 244         rcu_read_lock();
 245
 246         /* Find the vhost_vsock according to guest context id  */
 247         vsock = vhost_vsock_get(vsk->remote_addr.svm_cid);
 248         if (!vsock)
 249                 goto out;
 250
 251         spin_lock_bh(&vsock->send_pkt_list_lock);
 252         list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) {
 253                 if (pkt->vsk != vsk)
 254                         continue;
 255                 list_move(&pkt->list, &freeme);
 256         }
 257         spin_unlock_bh(&vsock->send_pkt_list_lock);
 258
 259         list_for_each_entry_safe(pkt, n, &freeme, list) {
 260                 if (pkt->reply)
 261                         cnt++;
 262                 list_del(&pkt->list);
 263                 virtio_transport_free_pkt(pkt);
 264         }
 265
 266         if (cnt) {
 267                 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
 268                 int new_cnt;
 269
 270                 new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
 271                 if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num)
 272                         vhost_poll_queue(&tx_vq->poll);
 273         }
 274
 275         ret = 0;
 276 out:
 277         rcu_read_unlock();
 278         return ret;
 279 }
 280
 281 static struct virtio_vsock_pkt *
 282 vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq,
 283                       unsigned int out, unsigned int in)
 284 {
 285         struct virtio_vsock_pkt *pkt;
 286         struct iov_iter iov_iter;
 287         size_t nbytes;
 288         size_t len;
 289
 290         if (in != 0) {
 291                 vq_err(vq, "Expected 0 input buffers, got %u\n", in);
 292                 return NULL;
 293         }
 294
 295         pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
 296         if (!pkt)
 297                 return NULL;
 298
 299         len = iov_length(vq->iov, out);
 300         iov_iter_init(&iov_iter, WRITE, vq->iov, out, len);
 301
 302         nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
 303         if (nbytes != sizeof(pkt->hdr)) {
 304                 vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
 305                        sizeof(pkt->hdr), nbytes);
 306                 kfree(pkt);
 307                 return NULL;
 308         }
 309
 310         if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM)
 311                 pkt->len = le32_to_cpu(pkt->hdr.len);
 312
 313         /* No payload */
 314         if (!pkt->len)
 315                 return pkt;
 316
 317         /* The pkt is too big */
 318         if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
 319                 kfree(pkt);
 320                 return NULL;
 321         }
 322
 323         pkt->buf = kmalloc(pkt->len, GFP_KERNEL);
 324         if (!pkt->buf) {
 325                 kfree(pkt);
 326                 return NULL;
 327         }
 328
 329         nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter);
 330         if (nbytes != pkt->len) {
 331                 vq_err(vq, "Expected %u byte payload, got %zu bytes\n",
 332                        pkt->len, nbytes);
 333                 virtio_transport_free_pkt(pkt);
 334                 return NULL;
 335         }
 336
 337         return pkt;
 338 }
 339
 340 /* Is there space left for replies to rx packets? */
 341 static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
 342 {
 343         struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
 344         int val;
 345
 346         smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
 347         val = atomic_read(&vsock->queued_replies);
 348
 349         return val < vq->num;
 350 }
 351
 352 static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
 353 {
 354         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 355                                                   poll.work);
 356         struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
 357                                                  dev);
 358         struct virtio_vsock_pkt *pkt;
 359         int head;
 360         unsigned int out, in;
 361         bool added = false;
 362
 363         mutex_lock(&vq->mutex);
 364
 365         if (!vq->private_data)
 366                 goto out;
 367
 368         vhost_disable_notify(&vsock->dev, vq);
 369         for (;;) {
 370                 u32 len;
 371
 372                 if (!vhost_vsock_more_replies(vsock)) {
 373                         /* Stop tx until the device processes already
 374                          * pending replies.  Leave tx virtqueue
 375                          * callbacks disabled.
 376                          */
 377                         goto no_more_replies;
 378                 }
 379
 380                 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
 381                                          &out, &in, NULL, NULL);
 382                 if (head < 0)
 383                         break;
 384
 385                 if (head == vq->num) {
 386                         if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
 387                                 vhost_disable_notify(&vsock->dev, vq);
 388                                 continue;
 389                         }
 390                         break;
 391                 }
 392
 393                 pkt = vhost_vsock_alloc_pkt(vq, out, in);
 394                 if (!pkt) {
 395                         vq_err(vq, "Faulted on pkt\n");
 396                         continue;
 397                 }
 398
 399                 len = pkt->len;
 400
 401                 /* Only accept correctly addressed packets */
 402                 if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid &&
 403                     le64_to_cpu(pkt->hdr.dst_cid) ==
 404                     vhost_transport_get_local_cid())
 405                         virtio_transport_recv_pkt(pkt);
 406                 else
 407                         virtio_transport_free_pkt(pkt);
 408
 409                 vhost_add_used(vq, head, sizeof(pkt->hdr) + len);
 410                 added = true;
 411         }
 412
 413 no_more_replies:
 414         if (added)
 415                 vhost_signal(&vsock->dev, vq);
 416
 417 out:
 418         mutex_unlock(&vq->mutex);
 419 }
 420
 421 static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
 422 {
 423         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 424                                                 poll.work);
 425         struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
 426                                                  dev);
 427
 428         vhost_transport_do_send_pkt(vsock, vq);
 429 }
 430
 431 static int vhost_vsock_start(struct vhost_vsock *vsock)
 432 {
 433         struct vhost_virtqueue *vq;
 434         size_t i;
 435         int ret;
 436
 437         mutex_lock(&vsock->dev.mutex);
 438
 439         ret = vhost_dev_check_owner(&vsock->dev);
 440         if (ret)
 441                 goto err;
 442
 443         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 444                 vq = &vsock->vqs[i];
 445
 446                 mutex_lock(&vq->mutex);
 447
 448                 if (!vhost_vq_access_ok(vq)) {
 449                         ret = -EFAULT;
 450                         goto err_vq;
 451                 }
 452
 453                 if (!vq->private_data) {
 454                         vq->private_data = vsock;
 455                         ret = vhost_vq_init_access(vq);
 456                         if (ret)
 457                                 goto err_vq;
 458                 }
 459
 460                 mutex_unlock(&vq->mutex);
 461         }
 462
 463         /* Some packets may have been queued before the device was started,
 464          * let's kick the send worker to send them.
 465          */
 466         vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
 467
 468         mutex_unlock(&vsock->dev.mutex);
 469         return 0;
 470
 471 err_vq:
 472         vq->private_data = NULL;
 473         mutex_unlock(&vq->mutex);
 474
 475         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 476                 vq = &vsock->vqs[i];
 477
 478                 mutex_lock(&vq->mutex);
 479                 vq->private_data = NULL;
 480                 mutex_unlock(&vq->mutex);
 481         }
 482 err:
 483         mutex_unlock(&vsock->dev.mutex);
 484         return ret;
 485 }
 486
 487 static int vhost_vsock_stop(struct vhost_vsock *vsock)
 488 {
 489         size_t i;
 490         int ret;
 491
 492         mutex_lock(&vsock->dev.mutex);
 493
 494         ret = vhost_dev_check_owner(&vsock->dev);
 495         if (ret)
 496                 goto err;
 497
 498         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 499                 struct vhost_virtqueue *vq = &vsock->vqs[i];
 500
 501                 mutex_lock(&vq->mutex);
 502                 vq->private_data = NULL;
 503                 mutex_unlock(&vq->mutex);
 504         }
 505
 506 err:
 507         mutex_unlock(&vsock->dev.mutex);
 508         return ret;
 509 }
 510
 511 static void vhost_vsock_free(struct vhost_vsock *vsock)
 512 {
 513         kvfree(vsock);
 514 }
 515
 516 static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
 517 {
 518         struct vhost_virtqueue **vqs;
 519         struct vhost_vsock *vsock;
 520         int ret;
 521
 522         /* This struct is large and allocation could fail, fall back to vmalloc
 523          * if there is no other way.
 524          */
 525         vsock = kzalloc(sizeof(*vsock), GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
 526         if (!vsock) {
 527                 vsock = vmalloc(sizeof(*vsock));
 528                 if (!vsock)
 529                         return -ENOMEM;
 530         }
 531
 532         vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
 533         if (!vqs) {
 534                 ret = -ENOMEM;
 535                 goto out;
 536         }
 537
 538         vsock->guest_cid = 0; /* no CID assigned yet */
 539
 540         atomic_set(&vsock->queued_replies, 0);
 541
 542         vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
 543         vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
 544         vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
 545         vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
 546
 547         vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs),
 548                        VHOST_VSOCK_PKT_WEIGHT,
 549                        VHOST_VSOCK_WEIGHT);
 550
 551         file->private_data = vsock;
 552         spin_lock_init(&vsock->send_pkt_list_lock);
 553         INIT_LIST_HEAD(&vsock->send_pkt_list);
 554         vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
 555         return 0;
 556
 557 out:
 558         vhost_vsock_free(vsock);
 559         return ret;
 560 }
 561
 562 static void vhost_vsock_flush(struct vhost_vsock *vsock)
 563 {
 564         int i;
 565
 566         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++)
 567                 if (vsock->vqs[i].handle_kick)
 568                         vhost_poll_flush(&vsock->vqs[i].poll);
 569         vhost_work_flush(&vsock->dev, &vsock->send_pkt_work);
 570 }
 571
 572 static void vhost_vsock_reset_orphans(struct sock *sk)
 573 {
 574         struct vsock_sock *vsk = vsock_sk(sk);
 575
 576         /* vmci_transport.c doesn't take sk_lock here either.  At least we're
 577          * under vsock_table_lock so the sock cannot disappear while we're
 578          * executing.
 579          */
 580
 581         /* If the peer is still valid, no need to reset connection */
 582         if (vhost_vsock_get(vsk->remote_addr.svm_cid))
 583                 return;
 584
 585         /* If the close timeout is pending, let it expire.  This avoids races
 586          * with the timeout callback.
 587          */
 588         if (vsk->close_work_scheduled)
 589                 return;
 590
 591         sock_set_flag(sk, SOCK_DONE);
 592         vsk->peer_shutdown = SHUTDOWN_MASK;
 593         sk->sk_state = SS_UNCONNECTED;
 594         sk->sk_err = ECONNRESET;
 595         sk->sk_error_report(sk);
 596 }
 597
 598 static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
 599 {
 600         struct vhost_vsock *vsock = file->private_data;
 601
 602         spin_lock_bh(&vhost_vsock_lock);
 603         if (vsock->guest_cid)
 604                 hash_del_rcu(&vsock->hash);
 605         spin_unlock_bh(&vhost_vsock_lock);
 606
 607         /* Wait for other CPUs to finish using vsock */
 608         synchronize_rcu();
 609
 610         /* Iterating over all connections for all CIDs to find orphans is
 611          * inefficient.  Room for improvement here. */
 612         vsock_for_each_connected_socket(vhost_vsock_reset_orphans);
 613
 614         vhost_vsock_stop(vsock);
 615         vhost_vsock_flush(vsock);
 616         vhost_dev_stop(&vsock->dev);
 617
 618         spin_lock_bh(&vsock->send_pkt_list_lock);
 619         while (!list_empty(&vsock->send_pkt_list)) {
 620                 struct virtio_vsock_pkt *pkt;
 621
 622                 pkt = list_first_entry(&vsock->send_pkt_list,
 623                                 struct virtio_vsock_pkt, list);
 624                 list_del_init(&pkt->list);
 625                 virtio_transport_free_pkt(pkt);
 626         }
 627         spin_unlock_bh(&vsock->send_pkt_list_lock);
 628
 629         vhost_dev_cleanup(&vsock->dev, false);
 630         kfree(vsock->dev.vqs);
 631         vhost_vsock_free(vsock);
 632         return 0;
 633 }
 634
 635 static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
 636 {
 637         struct vhost_vsock *other;
 638
 639         /* Refuse reserved CIDs */
 640         if (guest_cid <= VMADDR_CID_HOST ||
 641             guest_cid == U32_MAX)
 642                 return -EINVAL;
 643
 644         /* 64-bit CIDs are not yet supported */
 645         if (guest_cid > U32_MAX)
 646                 return -EINVAL;
 647
 648         /* Refuse if CID is already in use */
 649         spin_lock_bh(&vhost_vsock_lock);
 650         other = vhost_vsock_get(guest_cid);
 651         if (other && other != vsock) {
 652                 spin_unlock_bh(&vhost_vsock_lock);
 653                 return -EADDRINUSE;
 654         }
 655
 656         if (vsock->guest_cid)
 657                 hash_del_rcu(&vsock->hash);
 658
 659         vsock->guest_cid = guest_cid;
 660         hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid);
 661         spin_unlock_bh(&vhost_vsock_lock);
 662
 663         return 0;
 664 }
 665
 666 static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
 667 {
 668         struct vhost_virtqueue *vq;
 669         int i;
 670
 671         if (features & ~VHOST_VSOCK_FEATURES)
 672                 return -EOPNOTSUPP;
 673
 674         mutex_lock(&vsock->dev.mutex);
 675         if ((features & (1 << VHOST_F_LOG_ALL)) &&
 676             !vhost_log_access_ok(&vsock->dev)) {
 677                 mutex_unlock(&vsock->dev.mutex);
 678                 return -EFAULT;
 679         }
 680
 681         for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
 682                 vq = &vsock->vqs[i];
 683                 mutex_lock(&vq->mutex);
 684                 vq->acked_features = features;
 685                 mutex_unlock(&vq->mutex);
 686         }
 687         mutex_unlock(&vsock->dev.mutex);
 688         return 0;
 689 }
 690
 691 static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
 692                                   unsigned long arg)
 693 {
 694         struct vhost_vsock *vsock = f->private_data;
 695         void __user *argp = (void __user *)arg;
 696         u64 guest_cid;
 697         u64 features;
 698         int start;
 699         int r;
 700
 701         switch (ioctl) {
 702         case VHOST_VSOCK_SET_GUEST_CID:
 703                 if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
 704                         return -EFAULT;
 705                 return vhost_vsock_set_cid(vsock, guest_cid);
 706         case VHOST_VSOCK_SET_RUNNING:
 707                 if (copy_from_user(&start, argp, sizeof(start)))
 708                         return -EFAULT;
 709                 if (start)
 710                         return vhost_vsock_start(vsock);
 711                 else
 712                         return vhost_vsock_stop(vsock);
 713         case VHOST_GET_FEATURES:
 714                 features = VHOST_VSOCK_FEATURES;
 715                 if (copy_to_user(argp, &features, sizeof(features)))
 716                         return -EFAULT;
 717                 return 0;
 718         case VHOST_SET_FEATURES:
 719                 if (copy_from_user(&features, argp, sizeof(features)))
 720                         return -EFAULT;
 721                 return vhost_vsock_set_features(vsock, features);
 722         default:
 723                 mutex_lock(&vsock->dev.mutex);
 724                 r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
 725                 if (r == -ENOIOCTLCMD)
 726                         r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
 727                 else
 728                         vhost_vsock_flush(vsock);
 729                 mutex_unlock(&vsock->dev.mutex);
 730                 return r;
 731         }
 732 }
 733
 734 static const struct file_operations vhost_vsock_fops = {
 735         .owner          = THIS_MODULE,
 736         .open           = vhost_vsock_dev_open,
 737         .release        = vhost_vsock_dev_release,
 738         .llseek         = noop_llseek,
 739         .unlocked_ioctl = vhost_vsock_dev_ioctl,
 740 };
 741
 742 static struct miscdevice vhost_vsock_misc = {
 743         .minor = MISC_DYNAMIC_MINOR,
 744         .name = "vhost-vsock",
 745         .fops = &vhost_vsock_fops,
 746 };
 747
 748 static struct virtio_transport vhost_transport = {
 749         .transport = {
 750                 .get_local_cid            = vhost_transport_get_local_cid,
 751
 752                 .init                     = virtio_transport_do_socket_init,
 753                 .destruct                 = virtio_transport_destruct,
 754                 .release                  = virtio_transport_release,
 755                 .connect                  = virtio_transport_connect,
 756                 .shutdown                 = virtio_transport_shutdown,
 757                 .cancel_pkt               = vhost_transport_cancel_pkt,
 758
 759                 .dgram_enqueue            = virtio_transport_dgram_enqueue,
 760                 .dgram_dequeue            = virtio_transport_dgram_dequeue,
 761                 .dgram_bind               = virtio_transport_dgram_bind,
 762                 .dgram_allow              = virtio_transport_dgram_allow,
 763
 764                 .stream_enqueue           = virtio_transport_stream_enqueue,
 765                 .stream_dequeue           = virtio_transport_stream_dequeue,
 766                 .stream_has_data          = virtio_transport_stream_has_data,
 767                 .stream_has_space         = virtio_transport_stream_has_space,
 768                 .stream_rcvhiwat          = virtio_transport_stream_rcvhiwat,
 769                 .stream_is_active         = virtio_transport_stream_is_active,
 770                 .stream_allow             = virtio_transport_stream_allow,
 771
 772                 .notify_poll_in           = virtio_transport_notify_poll_in,
 773                 .notify_poll_out          = virtio_transport_notify_poll_out,
 774                 .notify_recv_init         = virtio_transport_notify_recv_init,
 775                 .notify_recv_pre_block    = virtio_transport_notify_recv_pre_block,
 776                 .notify_recv_pre_dequeue  = virtio_transport_notify_recv_pre_dequeue,
 777                 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
 778                 .notify_send_init         = virtio_transport_notify_send_init,
 779                 .notify_send_pre_block    = virtio_transport_notify_send_pre_block,
 780                 .notify_send_pre_enqueue  = virtio_transport_notify_send_pre_enqueue,
 781                 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
 782
 783                 .set_buffer_size          = virtio_transport_set_buffer_size,
 784                 .set_min_buffer_size      = virtio_transport_set_min_buffer_size,
 785                 .set_max_buffer_size      = virtio_transport_set_max_buffer_size,
 786                 .get_buffer_size          = virtio_transport_get_buffer_size,
 787                 .get_min_buffer_size      = virtio_transport_get_min_buffer_size,
 788                 .get_max_buffer_size      = virtio_transport_get_max_buffer_size,
 789         },
 790
 791         .send_pkt = vhost_transport_send_pkt,
 792 };
 793
 794 static int __init vhost_vsock_init(void)
 795 {
 796         int ret;
 797
 798         ret = vsock_core_init(&vhost_transport.transport);
 799         if (ret < 0)
 800                 return ret;
 801         return misc_register(&vhost_vsock_misc);
 802 };
 803
 804 static void __exit vhost_vsock_exit(void)
 805 {
 806         misc_deregister(&vhost_vsock_misc);
 807         vsock_core_exit();
 808 };
 809
 810 module_init(vhost_vsock_init);
 811 module_exit(vhost_vsock_exit);
 812 MODULE_LICENSE("GPL v2");
 813 MODULE_AUTHOR("Asias He");
 814 MODULE_DESCRIPTION("vhost transport for vsock ");